AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

WorldCompass: Reinforcement Learning for Long-Horizon World Models

Zehan Wang, Tengfei Wang, Haiyu Zhang, Xuhui Zuo, Junta Wu, Haoyuan Wang, Wenqiang Sun, Zhenwei Wang, Chenjie Cao, Hengshuang Zhao, Chunchao Guo, Zhou Zhao

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply WorldCompass RL post-trainingHY-Video-1.5 · WorldPlay long-term, Combined Actionaction accuracy: 19.70 -> 54.82Improved
apply WorldCompass RL post-trainingWan2.2 · WorldPlay long-term, Combined Actionaction accuracy: 19.58 -> 42.92Improved
apply WorldCompass RL post-trainingHY-Video-1.5 · WorldPlay short-term, Combined Actionaction accuracy: 21.74 -> 58.20Improved
apply WorldCompass RL post-trainingHY-Video-1.5 · WorldPlay long-term, Basic Actionaction accuracy: 64.28 -> 76.56Improved
apply WorldCompass RL post-trainingHY-Video-1.5 · WorldPlay long-term, Combined ActionHPSv3: -0.33 -> 0.73Improved
use clip-level instead of sample-level rollout (ablation)WorldCompass (HY-Video-1.5) · WorldPlay long-term, Combined Actionaction accuracy: 12.45 -> 54.82Improved
add visual quality reward (ablation)WorldCompass (HY-Video-1.5) · WorldPlay long-term, Combined ActionHPSv3: -2.67 -> 0.73Improved
add visual quality reward (ablation)WorldCompass (HY-Video-1.5) · WorldPlay long-term, Combined Actionaction accuracy: 36.39 -> 54.82Improved
add interaction following reward (ablation)WorldCompass (HY-Video-1.5) · WorldPlay long-term, Combined Actionaction accuracy: 11.51 -> 54.82Improved
add interaction following reward (ablation)WorldCompass (HY-Video-1.5) · WorldPlay long-term, Combined ActionHPSv3: 1.01 -> 0.73Hurt
use DiffusionNFT instead of DanceGRPO (ablation)WorldCompass (HY-Video-1.5) · WorldPlay long-term, Combined Actionaction accuracy: 20.02 -> 54.82Improved
add best-of-N sampling (ablation)WorldCompass (HY-Video-1.5) · WorldPlay long-term, Combined Actionaction accuracy: 55.28 -> 54.82No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy